Start a new topic

Bad Image detection/tracking quality after switching from .wtc to .zip

I was using .wtc file as target collection, but recently I switched to .zip because needed to generate targets runtime. Everything works as excepted but Image tracking quality is significantly worse compared to .wtc. Fake image detection  events are raised by SDK at random localtions. these detections lasts  about 0.2 - 1 second (after that, target lost event are raised). I noticed that during that short random detections augmentations are drawn very small (about 5 -10x smaller compared to when they are drawn when image target is detected correctly), so I assume that Wikitude  thinks that image target can be very small or it can be too far away.  Tried several different options to solve problem: disabled range extension, also used only 1 target image during testing... but result were same.


so I have several questions:

  • In case of .wtc, I had option to specify real size of image target in cm, is there way to specify image targets size in case of .zip?
  • I there any way to configure maximum range of detection?
    (for example, I don't want to detect image if it is more than 50cm far away from camera)
  • Is there any way to configure minimum size of detected image relative to camera viewport?
    (for example, I don't want to detect image if it not take more than 1/4 of camera viewport)


Environment

  • Unity Version - 2018
  • Wikitude SDK version -  Professional 9.8
  • Image target description - Bank credit card
  • Wikitude Arcore/Arkit integration disabled
1 Comment

Hi, 


I'm sorry to hear that you're having issues with the zip collections. What is the pixel size and image format of the target included in the zip? If possible, it would be great if you could share the .zip and .wtc you have as well, so that we can run the comparison internally. If those cannot be shared in the forums, you can also send them at support@wikitude.com, mentioning this forum ticket and I'll follow up there.


As for your questions, there is a way to set the image targets in the .zip collection by writing a .metadata file in the zip as well, however I will point out that this doesn't affect how recognition performs. It's only used in some APIs to compute the real distance between the camera and targets. Here is an example of such a file, with the "height" being specified in mm:



{
    "version": "1",
    "targets": [
    {
        "type": "Plane",
        "path": "holo_farmer.jpg",
        "name": "holo_farmer",
        "height": 148
    },
    {
        "type": "Plane",
        "path": "holo_flower.jpg",
        "name": "holo_flower",
        "height": 148
    }
    ]
}


There is no way at the moment to prefilter the recognition responses that you're getting from the SDK, but for now I think the focus should be to get the recognition and tracking quality of the zip collection to be comparable to the wtc version.


Best regards,

Alexandru

Login or Signup to post a comment